From dc51575a5f813164e96c7c1ab41aaf33585573f6 Mon Sep 17 00:00:00 2001
From: Eli Friedman <eli.friedman@gmail.com>
Date: Thu, 19 May 2011 22:16:13 +0000
Subject: [PATCH] Add fast-isel support for zeroext and signext ret
 instructions on x86.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131689 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/X86/X86FastISel.cpp        | 30 +++++++++++++++++----
 test/CodeGen/X86/fast-isel-ret-ext.ll | 38 +++++++++++++++++++++++++++
 2 files changed, 63 insertions(+), 5 deletions(-)
 create mode 100644 test/CodeGen/X86/fast-isel-ret-ext.ll

diff --git a/lib/Target/X86/X86FastISel.cpp b/lib/Target/X86/X86FastISel.cpp
index a82db891761..6e96638c605 100644
--- a/lib/Target/X86/X86FastISel.cpp
+++ b/lib/Target/X86/X86FastISel.cpp
@@ -722,18 +722,38 @@ bool X86FastISel::X86SelectRet(const Instruction *I) {
     // Only handle register returns for now.
     if (!VA.isRegLoc())
       return false;
-    // TODO: For now, don't try to handle cases where getLocInfo()
-    // says Full but the types don't match.
-    if (TLI.getValueType(RV->getType()) != VA.getValVT())
-      return false;
 
     // The calling-convention tables for x87 returns don't tell
     // the whole story.
     if (VA.getLocReg() == X86::ST0 || VA.getLocReg() == X86::ST1)
       return false;
 
-    // Make the copy.
     unsigned SrcReg = Reg + VA.getValNo();
+    EVT SrcVT = TLI.getValueType(RV->getType());
+    EVT DstVT = VA.getValVT();
+    // Special handling for extended integers.
+    if (SrcVT != DstVT) {
+      if (SrcVT != MVT::i1 && SrcVT != MVT::i8 && SrcVT != MVT::i16)
+        return false;
+
+      if (!Outs[0].Flags.isZExt() && !Outs[0].Flags.isSExt())
+        return false;
+
+      assert(DstVT == MVT::i32 && "X86 should always ext to i32");
+
+      if (SrcVT == MVT::i1) {
+        if (Outs[0].Flags.isSExt())
+          return false;
+        SrcReg = FastEmitZExtFromI1(MVT::i8, SrcReg, /*TODO: Kill=*/false);
+        SrcVT = MVT::i8;
+      }
+      unsigned Op = Outs[0].Flags.isZExt() ? ISD::ZERO_EXTEND :
+                                             ISD::SIGN_EXTEND;
+      SrcReg = FastEmit_r(SrcVT.getSimpleVT(), DstVT.getSimpleVT(), Op,
+                          SrcReg, /*TODO: Kill=*/false);
+    }
+
+    // Make the copy.
     unsigned DstReg = VA.getLocReg();
     const TargetRegisterClass* SrcRC = MRI.getRegClass(SrcReg);
     // Avoid a cross-class copy. This is very unlikely.
diff --git a/test/CodeGen/X86/fast-isel-ret-ext.ll b/test/CodeGen/X86/fast-isel-ret-ext.ll
new file mode 100644
index 00000000000..fd768cb9b33
--- /dev/null
+++ b/test/CodeGen/X86/fast-isel-ret-ext.ll
@@ -0,0 +1,38 @@
+; RUN: llc < %s  -O0 -fast-isel-abort -mtriple i686-apple-darwin10 | FileCheck %s
+; RUN: llc < %s  -O0 -fast-isel-abort -mtriple x86_64-apple-darwin10 | FileCheck %s
+
+define zeroext i8 @test1(i32 %y) nounwind {
+  %conv = trunc i32 %y to i8
+  ret i8 %conv
+  ; CHECK: test1:
+  ; CHECK: movzbl {{.*}}, %eax
+}
+
+define signext i8 @test2(i32 %y) nounwind {
+  %conv = trunc i32 %y to i8
+  ret i8 %conv
+  ; CHECK: test2:
+  ; CHECK: movsbl {{.*}}, %eax
+}
+
+define zeroext i16 @test3(i32 %y) nounwind {
+  %conv = trunc i32 %y to i16
+  ret i16 %conv
+  ; CHECK: test3:
+  ; CHECK: movzwl {{.*}}, %eax
+}
+
+define signext i16 @test4(i32 %y) nounwind {
+  %conv = trunc i32 %y to i16
+  ret i16 %conv
+  ; CHECK: test4:
+  ; CHECK: movswl {{.*}}, %eax
+}
+
+define zeroext i1 @test5(i32 %y) nounwind {
+  %conv = trunc i32 %y to i1
+  ret i1 %conv
+  ; CHECK: test5:
+  ; CHECK: andb $1
+  ; CHECK: movzbl {{.*}}, %eax
+}